Is It Actually Possible to Overcome Our Prejudices?

Prejudice, a derogatory attitude or antipathy based upon unfair generalization, runs deeply in the human psyche. As a seminal review noted, “prejudice is animus, or negative bias, toward social groups and their putative members”.

However you feel about prejudice, there’s no questioning its ubiquity in society, by which I mean any human collective.

Anyone who claims otherwise is either an incorrigible optimist, or merely ignorant in psychological terms; but I may be prejudiced…

OK, so let’s look at some frequently asked questions…

How many people are prejudiced?

Pretty much everyone, including (perhaps especially) the people who loudly tell you they are not. Actually, let me rephrase: absolutely everyone – to be human = to be biased by design. If reading this annoys you, you can blame overconfident self-deception, the mother of all biases.

Decades of research on implicit social cognition show that human adults display automatic preferences for high-status or majority groups, even when they endorse egalitarian values. Greenwald and Banaji’s classic work on implicit attitudes showed that stereotyping is an ordinary byproduct of normal learning and categorization, not a pathology found in a few bad apples. Large-scale data from Project Implicit, which includes more than 2.3 million tests from 34 countries, backs this up.

These effects are not limited to members of dominant groups. Members of disadvantaged groups often show the same pro-dominant implicit pattern, which suggests that prejudice reflects internalized cultural hierarchies rather than just personal animus.

So, the question is not “who is prejudiced” but “about whom, in what situations, and how strongly.” From a scientific standpoint, a completely prejudice-free human would be an outlier, perhaps even a statistical error, if not an AI agent… (but I’ll get to that in question 6 below).

1. Why does prejudice exist?

Prejudice is not a software bug but a default feature of human cognition that becomes dangerous when it goes unexamined or is amplified by power. Three ingredients are well supported, namely:

Cognitive efficiency. The brain is a compulsive categorization machine. Stereotypes economize thinking by compressing complex information into group labels, which is helpful for quick decisions yet terrible for fairness. Work on implicit social cognition shows how these learned associations are triggered automatically and outside awareness.

Motivations to manage threat and hierarchy. Social dominance theory argues that many people are motivated to preserve group-based hierarchies, which leads to prejudice against groups seen as lower status or socially disruptive.

Norms and institutions. Meta-analytic work on generalized prejudice shows that these ideological motives are shaped by the social environment: when inequality is high and group hierarchies are legitimized, both social dominance orientation and right-wing authoritarianism rise and so does prejudice.

Thus, prejudice exists because it once solved adaptive problems (who is “us,” who is “them,” who is safe) and is continually reinforced by cultural narratives and institutional incentives.

2. Are some people more prejudiced than others, and if so, why?

Yes. Prejudice is a human universal, but not evenly distributed. Some people are reliably more biased across many target groups and contexts, and we know quite a bit about why.

Meta-analyses show that people high in right-wing authoritarianism, who value conformity and authority, or high in social dominance orientation, who are comfortable with group-based hierarchies, tend to oppose equality-enhancing policies and hold negative views of low status groups.

So you get a consistent profile of the chronically prejudiced person: low openness, low empathy, strong preference for hierarchy and order, and a worldview in which the world is dangerous and “people like us” must keep “people like them” in their place.

3. How effective are prejudice-reduction interventions?

The evidence says that prejudice can be reduced, but not easily, not quickly, and not by a two-hour workshop sandwiched between quarterly results and the cheese platter.

A major review of 418 experiments concluded that many interventions have small to moderate effects on attitudes, but impact on actual behavior is often weaker and short-lived. Intergroup contact remains one of the most reliable tools. By contrast, a recent review of bias and diversity trainings warns that one-off, non-scientific workshops frequently show no effect or even increase defensiveness and bias.

Bias Essential Reads

The Hidden Costs of Anti-Bias Education

When the Brain Shapes Belief

4. How can you find out your own level of prejudice?

There is no MRI or implant test for the racist brain. What we have are imperfect, probabilistic indicators that are more informative in combination than in isolation.

Implicit measures. The Implicit Association Test (IAT) is the most famous. Meta-analytic work finds a small to moderate correlation, around r = .24, between implicit and explicit attitudes (that means less than 6% of overlap!).

Explicit self-report scales. Modern forms of prejudice are often captured by subtle measures such as symbolic racism scales, attitudes toward immigration, or support for group-based hierarchy. These predict policy preferences and voting behavior at least as well, and often better, than implicit measures.

Behavioral data. For workplaces, the most honest prejudice test is your own track record. Who do you hire, promote, sponsor or invite to stretch assignments, after controlling for objective performance indicators? Studies of bias in healthcare, education and hiring show that even small average biases at decision points can compound into large inequalities over time.

5. Is AI less prejudiced than humans, and will it make humans more or less prejudiced?

The short answer: AI is neither the cure for human prejudice nor its inevitable amplifier. Whether it becomes more or less biased than us depends almost entirely on the humans who build, deploy and monitor it. In other words, AI does not transcend human prejudice; it scales it.

First, the optimistic view. Unlike humans, AI systems do not have emotions, status anxieties, ingroup loyalty or evolutionary threat detectors. This means that, in principle, we can audit, quantify and correct algorithmic biases far more systematically than we can rewire human cognition.

But the optimistic story has a rather large asterisk: most AI tools are trained on human data, and human data is biased. So, the more realistic conclusion is this: AI is not inherently less prejudiced than humans; it simply reflects the statistical patterns we feed it.

6. What does all this mean for workplaces?

The science is quite sobering. Prejudice is widespread. Some people are more biased than others, and a few very biased people in positions of power can do a lot of damage. Interventions can help, but mostly when they are intensive, sustained and embedded in structures and norms, not when they are outsourced to a glossy vendor and ticked off as “completed”.

In other words, you are unlikely to eradicate prejudice from people’s minds, including your own. What you can do is make sure those prejudices have fewer chances to quietly shape who gets hired, heard, developed or fired.